Archived from groups: rec.games.frp.gurps (More info?)
> >> you can keep what they are able to learn
> >> under control by out-of-game factors, like the amount of Modular
> >> Abilities they have.
> >
> >No, modular abilities do work nicely to model standard computer programs
> >running on today's machines, not the TL9 learning architectures of
> >Transhuman Space's computers. Having infomorphs use some kind of slots to
> >use skills, with limited memory and the like, would strongly go against
the
> >setting, which clearly sets a difference between software to run and
learned
> >skills.
>
> I know NOTHING of the setting, so let me know: these machines are able
> to record EVERYTHING and to learn as fast as they can download?
> And to use it at once, without delay, every time it is needed?
>
Of course not, information needs to be processed and, in order to generate a
skill increase, the machine, just like a human, must spend time 'learning'
it, using todays AI programming concepts to describe it I would say that
their clouds of actors need time to evolve, that their software neural nets
reconfigure and develop new strategies. The setting presumes that Sapient
AIs are capable of thought just like humans do, and that Non-sapient and
Low-sapient AIs are still able to learn stuff much like humans do, just from
experience, without pre-coded algorithms.
Once the AI has learnt it though, it can use the new skill every time it
needs, just as a trained human brain can form meaningful words every time it
needs, one it has learnt how to speak.
> It seems thousand of points worth, if it is so.
It depends on how fast raw data results in learning. That being the starting
point of the discussion in fact.
> I mean, in this setting teaching is dead?
No, of course not, just yesterday I was sitting at my computer, writing a
little program which basically will propose a few millions scenarios to a
genetic algorithm that I'll prepare next week, I guess I'll have it run for
a couple minutes, I hope it (the neural net) will learn something useful,
namely how to sort data more efficiently. That's teaching, and it may take a
while. TS take on computers is much the same as today's artificial life
algorithms, so teaching is not dead at all, just as today's AI software, it
takes a good 'teacher' to have the AI learn something useful, and it takes
time. The problem is: how much time? ETS and the economics of AI software
proposed in Transhuman Space seem to suggest a relatively high rate of
learning on the part of AI operating systems. Which is a true mess when you
have PCs using AIs.
> >> you can keep what they are able to learn
> >> under control by out-of-game factors, like the amount of Modular
> >> Abilities they have.
> >
> >No, modular abilities do work nicely to model standard computer programs
> >running on today's machines, not the TL9 learning architectures of
> >Transhuman Space's computers. Having infomorphs use some kind of slots to
> >use skills, with limited memory and the like, would strongly go against
the
> >setting, which clearly sets a difference between software to run and
learned
> >skills.
>
> I know NOTHING of the setting, so let me know: these machines are able
> to record EVERYTHING and to learn as fast as they can download?
> And to use it at once, without delay, every time it is needed?
>
Of course not, information needs to be processed and, in order to generate a
skill increase, the machine, just like a human, must spend time 'learning'
it, using todays AI programming concepts to describe it I would say that
their clouds of actors need time to evolve, that their software neural nets
reconfigure and develop new strategies. The setting presumes that Sapient
AIs are capable of thought just like humans do, and that Non-sapient and
Low-sapient AIs are still able to learn stuff much like humans do, just from
experience, without pre-coded algorithms.
Once the AI has learnt it though, it can use the new skill every time it
needs, just as a trained human brain can form meaningful words every time it
needs, one it has learnt how to speak.
> It seems thousand of points worth, if it is so.
It depends on how fast raw data results in learning. That being the starting
point of the discussion in fact.
> I mean, in this setting teaching is dead?
No, of course not, just yesterday I was sitting at my computer, writing a
little program which basically will propose a few millions scenarios to a
genetic algorithm that I'll prepare next week, I guess I'll have it run for
a couple minutes, I hope it (the neural net) will learn something useful,
namely how to sort data more efficiently. That's teaching, and it may take a
while. TS take on computers is much the same as today's artificial life
algorithms, so teaching is not dead at all, just as today's AI software, it
takes a good 'teacher' to have the AI learn something useful, and it takes
time. The problem is: how much time? ETS and the economics of AI software
proposed in Transhuman Space seem to suggest a relatively high rate of
learning on the part of AI operating systems. Which is a true mess when you
have PCs using AIs.
Facebook
Twitter
Instagram